Is it memory, realistic dialogue, strong character personalities, unexpected reactions, or something else?
I am curious what separates a good AI roleplay conversation from one that starts feeling repetitive
Consider this a rant because I honestly do not know where else to put this frustration.
My third book is AI assisted, and that does not mean I cannot write. My first novel was self published in 2022, and every word of that 100,000 plus word manuscript was written and edited by me. I revised it more times than I can count, to the point where I almost stopped wanting to write altogether.
I started writing when I was fifteen. I was the quiet kid who spent more time in fictional worlds than the real one. Back then, I thought finishing a book was the hardest part. Nobody tells you that finishing the manuscript is actually where the real work begins.
Editing is not just fixing grammar. There are developmental edits, structural changes, line edits, rewrites, and proofreading. Professional editing can easily cost thousands of dollars, which is a huge barrier when you are not wealthy or connected to the publishing industry.
Then there is marketing.
You are suddenly expected to become a content creator, advertiser, strategist, and public personality while also working a normal job and trying to write your next book.
Post constantly. Build an audience. Learn algorithms. Make videos. Track engagement. Run ads. Stay visible.
It becomes exhausting very quickly.
That is why I started using AI tools. Not because I wanted them to write my books for me, but because I needed help managing everything around the actual writing.
Tasks that used to take months can now take weeks. I can catch structural problems faster, test different dialogue options, find plot holes, and get through revisions without spending an entire night staring at the same paragraph.
I still create the world. I create the characters. I decide their personalities, fears, motivations, relationships, and the direction of the story. The ideas come from me and my experiences.
AI is simply another tool in the process.
I do not believe writers need to suffer endlessly for their work to be considered authentic. I already know what burnout feels like, and I do not think destroying myself for every manuscript makes the final book more meaningful.
If someone wants to write every word completely alone, I respect that. But writers have always adapted to new tools.
Using technology to make the process more manageable does not make me less of a writer. I am still the person creating the story. I just refuse to believe suffering is a requirement for proving that I am one.
A polished version can be technically better but still feel strange when you read it out loud
Do you change the prompt to match your speaking style, or edit the response yourself?
Well, I definitely did.
I complained about seeing way too much AI generated content in a community that is supposed to be about creativity, writing novels, and discussing writing techniques.
Instead of getting a normal discussion, I got heavily downvoted and people acted like accusing someone of using AI was somehow worse than the AI content itself. My post eventually got removed for supposedly trying to start a witch hunt.
The funny part is that I was just looking for something new to read. I found a novel that kept getting recommended and had plenty of positive comments, but the writing immediately felt familiar.
The same "not A, not B, just C" structure showing up constantly, forced edgy one liners, flat one sentence dialogue, and characters that felt more like vague descriptions than actual people.
Maybe a human really did write it. I could be completely wrong about the AI part.
But even if that is the case, the writing still felt pretty bad to me. What confused me most was seeing almost everyone praise it as incredibly well written while criticism of the style was treated like some kind of attack.
It could ask a few questions first about the audience, purpose, tone, and desired result
The extra step might make conversations slower, but the first response could be much more relevant
Would you prefer that over getting an immediate draft?
When a request could mean several things, I ask the model to show how it understands the request before answering
That gives me a chance to correct the direction before getting a long response based on the wrong assumption
Has anyone else found this useful?
The grammar improves and everything flows nicely, but sometimes the personality disappears along with the mistakes
Do you usually keep the polished version or add your own wording back?
I've tried giving Al examples of my writing, but the results can still feel slightly off. Has anyone managed to get Al to consistently match their natural style?
One option keeps the workflow simple.
The other might produce better results for research, writing, and editing.
Which setup would you actually choose?
The explanation did not make much sense at first, but after trying it, the result was actually better than expected. Has that ever happened to you?
Not because it gave new facts, but because the discussion exposed a weakness in your own reasoning. Has that ever happened to you?
Sometimes the model points out a weak argument or confusing paragraph that I really wanted to keep. Has Al ever made you completely rethink something you had already finished?
Sometimes the feedback points out a problem that is impossible to ignore afterward. Has Al ever convinced you to remove something you originally thought was one of the strongest parts?
Imagine an Al that actively looked for flaws, disagreed with weak arguments, and refused to take your side without evidence. Would that make it more useful, or would it become frustrating over time?
If you had to give up one forever, which would be harder to replace in your daily workflow? I think the answers would be more divided than most people expect.
Everyone seems to have that one prompt they keep going back to. Whether it is for casual chats, work messages, or dating, I am curious to see the prompts people trust the most.
Sometimes a wrong answer sounds convincing because the model tries to be helpful. Would you rather see more honest uncertainty, even if it means getting fewer complete answers?